Quantifying Multipollutant Health Impacts Using the Environmental Benefits Mapping and Analysis Program–Community Edition (BenMAP-CE): A Case Study in Atlanta, Georgia

Abstract

Background:

Air pollution risk assessments do not generally quantify health impacts using multipollutant risk estimates, but instead use results from single-pollutant or copollutant models. Multipollutant epidemiological models account for pollutant interactions and joint effects but can be computationally complex and data intensive. Risk estimates from multipollutant studies are therefore challenging to implement in the quantification of health impacts.

Objectives:

Our objective was to conduct a case study using a developmental multipollutant version of the Environmental Benefits Mapping and Analysis Program—Community Edition (BenMAP-CE) to estimate the health impact associated with changes in multiple air pollutants using both a single and multipollutant approach.

Methods:

BenMAP-CE was used to estimate the change in the number of pediatric asthma emergency department (ED) visits attributable to simulated changes in air pollution between 2011 and 2025 in Atlanta, Georgia, applying risk estimates from an epidemiological study that examined short-term single-pollutant and multipollutant (with and without first-order interactions) exposures. Analyses examined individual pollutants (i.e., ozone, fine particulate matter, carbon monoxide, nitrogen dioxide (${\mathrm{NO}}_2$), sulfur dioxide, and particulate matter components) and combinations of these pollutants meant to represent shared properties or predefined sources (i.e., oxidant gases, secondary pollutants, traffic, power plant, and criteria pollutants). Comparisons were made between multipollutant health impact functions (HIF) and the sum of single-pollutant HIFs for the individual pollutants that constitute the respective pollutant groups.

Results:

Photochemical modeling predicted large decreases in most of the examined pollutant concentrations between 2011 and 2025 based on sector specific (i.e., source-based) estimates of growth and anticipated controls. Estimated number of avoided asthma ED visits attributable to any given multipollutant group were generally higher when using results from models that included interaction terms in comparison with those that did not. We estimated the greatest number of avoided pediatric asthma ED visits for pollutant groups that include ${\mathrm{NO}}_2$ (i. e., criteria pollutants, oxidants, and traffic pollutants). In models that accounted for interaction, year-round estimates for pollutant groups that included ${\mathrm{NO}}_2$ ranged from 27.1 [95% confidence interval (CI): 1.6, 52.7; traffic pollutants] to 55.4 (95% CI: 41.8, 69.0; oxidants) avoided pediatric asthma ED visits. Year-round results using multipollutant risk estimates with interaction were comparable to the sum of the single-pollutant results corresponding to most multipollutant groups [e.g., 52.9 (95% CI: 43.6, 62.2) for oxidants] but were notably lower than the sum of the single-pollutant results for some pollutant groups [e.g., 77.5 (95% CI: 66.0, 89.0) for traffic pollutants].

Discussion:

Performing a multipollutant health impact assessment is technically feasible but computationally complex. It requires time, resources, and detailed input parameters not commonly reported in air pollution epidemiological studies. Results estimated using the sum of single-pollutant models are comparable to those quantified using a multipollutant model. Although limited to a single study and location, assessing the trade-offs between a multipollutant and single-pollutant approach is warranted. https://doi.org/10.1289/EHP12969

Introduction

The Clean Air Act^1,2 requires the US Environmental Protection Agency (US EPA) to evaluate the scientific evidence to determine the health and welfare effects posed by criteria air pollutants [current criteria pollutants are particulate matter (PM), ozone (${\mathrm{O}}_3$), and related photochemical oxidants, oxides of nitrogen [e.g., nitrogen dioxide (${\mathrm{NO}}_2$)]; sulfur oxides, [e.g., sulfur dioxide (${\mathrm{SO}}_2$)]; carbon monoxide (CO); and lead (Pb)] to support the setting of National Ambient Air Quality Standards (NAAQS). Although it is well recognized that people are exposed to a complex mixture of air pollutants, the review of the scientific evidence and the subsequent establishment of each NAAQS have historically focused on conducting reviews for one pollutant at a time. The parameters of the Clean Air Act, which govern the review of each NAAQS separately, have contributed to researchers often examining the health effects of exposure to individual criteria pollutants, rather than simultaneous exposure to multiple pollutants.

The 2004 National Research Council (NRC) report “Air Quality Management in the United States” recommended that the US EPA transition from a “pollutant-by-pollutant” approach to managing air quality to a “multipollutant, risk-based approach.”³ More recent reports from the UK Committee on the Medical Effects of Air Pollutants⁴ and the National Academies of Sciences, Engineering and Medicine⁵ have similarly emphasized the need for a multipollutant approach to quantifying the health effects of air pollution exposures. Such a transition is largely predicated on the availability of epidemiological studies that account for multipollutant exposures and risk. Consistent with the NRC’s recommendation, numerous publications have provided additional calls for research aimed at understanding the health effects of multipollutant exposures while also acknowledging the many challenges of multipollutant research.^6–11 These calls for multipollutant research have led to the proliferation of novel multipollutant statistical approaches over the last 10 y.^12–15 Although these innovative methods have expanded our understanding of multipollutant exposures, multipollutant methods vary in complexity in relation to both model implementation and the interpretability of results,^16,17 which has complicated the comparison of results across studies.

An important downstream consideration when evaluating multipollutant epidemiological studies is the potential application of results from such studies in the quantification of the public health impacts and corresponding economic values of various air quality actions using tools such as US EPA’s Environmental Benefits and Mapping Program–Community Edition (BenMAP-CE), which is a free and open-source software program that employs user inputs to calculate air pollution-related health impacts.¹⁸ Risk assessments and regulatory impact analyses using BenMAP-CE, which quantitatively characterize air pollution risks, require knowledge of the relationship between air pollution exposure and a given health outcome, often represented by an effect estimate [also referred to as a risk estimate or beta ($\mathrm{\beta }$) coefficient] from an epidemiological study.¹⁸ To date, although some assessments have focused on risk-based multipollutant control strategies,¹⁹ health impact analyses have relied exclusively on results from epidemiological studies that represent the estimated health effect(s) related to variation in the concentration of a single pollutant while controlling for various confounders. Conversely, a multipollutant risk assessment strategy that evaluates the health effects related to a simultaneous change in multiple air pollutants represents an important but underdeveloped component of a multipollutant air quality management framework.²⁰ Implementation of such an approach to quantify health impacts of multiple air pollutants requires effect estimates from multipollutant epidemiological studies that are easily interpretable and able to be parameterized. However, increased dimensionality and collinearity in multipollutant studies often require more complex models to discern statistical relationships, which can decrease the interpretability of multipollutant effect estimates for the purposes of health impact analyses.²¹

In this case study, we estimate pediatric asthma emergency department (ED) visits attributable to multipollutant exposures using a developmental version of BenMAP-CE to demonstrate the application of a multipollutant assessment of health impacts, using effect estimates from an epidemiological study that modeled the joint effects of short-term multipollutant exposures on pediatric asthma ED visits in Atlanta, Georgia.²² In this analysis, we compare the procedures for quantifying population health impacts in a single and multipollutant context for Atlanta due to potential air quality changes between 2011 and 2025 resulting from sector specific (i.e., source-based) estimates of growth and anticipated controls.

Materials and Methods

Overview

This analysis used a developmental version of BenMAP-CE that has been modified to allow for the use of multipollutant health impact functions (HIFs) with and without interaction terms to estimate health impacts. BenMAP-CE is a US EPA tool that estimates the number and economic value of air pollution–related deaths and illnesses relying on preloaded or user-inputted datasets of air quality data, demographic data (i.e., baseline incidence and population data), concentration–response relationships for multiple individual air pollutants and numerous health outcomes, and economic values.²³

In this case study, we used air quality data predicted using the US EPA Community Multiscale Air Quality (CMAQ) model²⁴ for the years 2011 and a future year, 2025, based on sector specific (i.e., source-based) estimates of growth and anticipated controls to examine the relationship between changes in air quality in Atlanta and the estimated number of air pollution–related pediatric asthma ED visits. Health impacts were estimated in relation to predicted changes between 2011 and 2025 in pollutant concentrations using the baseline incidence of pediatric asthma ED visits in Atlanta, the corresponding population (i.e., number of children in Atlanta), and multipollutant concentration–response (C-R) functions from Winquist et al.²² relating joint exposure to groups of related air pollutants to asthma-related pediatric ED visits. Air pollutant groups identified a priori by Winquist et al.²² were also examined and included combinations of pollutants with shared properties (e.g., oxidant gases or secondary pollutants) and sources (e.g., traffic-related or coal-fired power plant pollutants) (Table 1). Finally, considering the computational burden of conducting multipollutant assessments, we compared results from the multipollutant HIFs to the sum of HIFs for the single pollutants that form the multipollutant groups to assess the utility of quantifying the health effects of air pollutants in a multipollutant context.

Table 1.

Pollutant groups and associated pollutants, as defined by Winquist et al.,²² used in the BenMAP-CE quantification of multipollutant impacts on asthma-related ED visits.

Pollutant group	Pollutants
Single pollutants	CO, EC, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, ${\mathrm{PM}}_{2.5}$, secondary ${\mathrm{PM}}_{2.5}$, ${\mathrm{SO}}_2$, and sulfate
Criteria pollutants	CO, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, ${\mathrm{PM}}_{2.5}$, and ${\mathrm{SO}}_2$
Oxidants	${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, and ${\mathrm{SO}}_2$
Power plant	${\mathrm{SO}}_2$ and sulfate
Secondary pollutants	${\mathrm{O}}_3$ and secondary ${\mathrm{PM}}_{2.5}$
Traffic pollutants	CO, EC, and ${\mathrm{NO}}_2$

Note: BenMAP-CE, Environmental Benefits Mapping and Analysis Program–Community Edition; CO, carbon monoxide; EC, elemental carbon; ED, emergency department; ${\mathrm{NO}}_2$, nitrogen dioxide, ${\mathrm{O}}_3$, ozone; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; ${\mathrm{SO}}_2$, sulfur dioxide.

Defining Multipollutant HIFs

As described in Sacks et al.,¹⁸ HIFs are used in BenMAP-CE to quantify the health impacts attributable to changes in an air pollutant (Equation 1). HIFs are commonly defined as:

$$\mathrm{\Delta }Y=\left(1-e^{-\mathrm{\beta }\times \mathrm{\Delta }AQ}\right)\times Y_0\times Pop,$$ (1)

where $\mathrm{\Delta }\mathrm{Y}$ is the estimated health impact attributable to a change in air quality ($\mathrm{\Delta }AQ$) in a given population (Pop) with a known baseline incidence ($Y_0$) for the health effect of interest. $\mathrm{\Delta }\mathrm{Y}$ is calculated using the effect estimate or beta coefficient ($\mathrm{\beta }$) from an epidemiological study.

Although HIFs have most commonly been used to estimate the health impacts attributable to a single air pollutant, the functional form is also able to accommodate multiple $\mathrm{\beta }$s from a multipollutant model that simultaneously adjusts for co-varying pollutants (Equation 2). In this analysis, we modified the HIF capabilities of BenMAP-CE to support the following form:

$$\mathrm{\Delta }Y=Y_0\left(1-e^{-\left({\sum }_{i=1}^n{\mathrm{\beta }}_i\times \mathrm{\Delta }A{Q}_i\right)}\right)\times Pop,$$ (2)

where $\mathrm{\Delta }\mathrm{Y}$ is the estimated joint health impact attributable to a simultaneous change in a set of $n$ air pollutants [e.g., ${\mathrm{\beta }}_i$ and $\mathrm{\Delta }A{Q}_i$ are the effect estimate and change in air quality for a given air pollutant (i) in a set of pollutants ($n$), respectively]. In addition, we extended the HIF in BenMAP-CE to support beta coefficients from multipollutant models with interaction terms (Equation 3), as follows:

$$\mathrm{\Delta }Y=Y_0\left(1-e^{-\left({\sum }_{i=1}^n{\mathrm{\beta }}_i\times \mathrm{\Delta }A{Q}_i+{\sum }_{j=1}^{n-1}{\sum }_{k=j+1}^n{\mathrm{\beta }}_{jk}\times \mathrm{\Delta }A{Q}_{jk}\right)}\right)\times Pop.$$ (3)

In this multipollutant model with interaction terms, ${\mathrm{\beta }}_{jk}$ represents the coefficient for the interaction terms for the given air pollutants (j and k) and $\mathrm{\Delta }A{Q}_{jk}$ is the difference in the products of the corresponding air pollutant concentrations, e.g., $\left[b_j\times b_k\right]-\left[a_j\times a_k\right]$, where b is the air pollutant concentration in 2025, and a is the air pollutant concentration in 2011.

The developmental multipollutant version of BenMAP-CE was modified to process HIFs in the forms described in Equations 2 and 3, in addition to the standard single-pollutant form highlighted in Equation 1. The data sources and preparation steps for the four HIF inputs ($\mathrm{\beta }$, $Y_0$, $\mathrm{\Delta }AQ$, and Pop) and standard error (SE) calculations are summarized in Table 2 and described in further detail in the ensuing sections.

Table 2.

Sources of input data corresponding to single and multipollutant health impact functions (as defined in Equations 1, 2, and 3) in the BenMAP-CE analysis.

Input	Source	Notes
Air quality (AQ)	Community Multiscale Air Quality model (CMAQ)	$12\times 12\mathrm{km}$ daily predicted concentrations for eight pollutants in 2011 and 2025
Baseline incidence of ED visits ($Y_0$)	County level incidence rates of pediatric asthma ED visits for children 2–18 y of age developed using data from the Healthcare Cost and Utilization Project (HCUP)14	Disaggregated to $12\times 12\mathrm{km}$ grid.
Population (Pop)	Population data was generated for children 5–17 y of age using 2010 US census block population data.	BenMAP-CE crosswalks to $12\times 12\mathrm{km}$ grid.
Beta coefficients (β)	Winquist et al.22	Single pollutant, joint effects, and joint effects with interaction

Note: BenMAP-CE, Environmental Benefits and Mapping Program–Community Edition; ED, emergency department; km, kilometer.

Epidemiological Data ($\mathrm{\beta }$ and SE)

The epidemiological data used in this analysis (Excel Table S1) were derived from a study²² that examined associations between short-term air pollution exposures, in both a single-pollutant and multipollutant context, and pediatric (i.e., children 5–17 y of age) asthma ED visits in Atlanta from 1998 to 2004. We used effect estimates from Winquist et al.²² for this case study because the statistical methodology used to examine multipollutant exposures was less computationally intensive than other multipollutant approaches in the literature and the results were more easily interpretable, because the reported joint effects represented a priori defined pollutant groups with shared properties or sources (i.e., pollutants that could be controlled simultaneously). The selected study was additionally advantageous because it included a health end point with extensive scientific evidence of a relationship with a number of pollutants and covered a limited geographic area, requiring less processing power from BenMAP-CE.

Winquist et al.²² reported associations using a multiday lag of 0–2 d (3-day moving average) for each air pollutant. With respect to multipollutant exposures, adjusted models were examined with and without first-order pollutant interactions for subsets of related air pollutants, with some groups representing source categories identified a priori²² (Table 1). Winquist et al.²² estimated rate ratios (RR) and 95% confidence intervals (CIs) from warm and cold season models for Atlanta. In addition to the multipollutant models, Winquist et al.²² also provided results from single-pollutant models for each of the air pollutants that comprise the multipollutant groups (Table 1). The single-pollutant models were not adjusted for copollutants, but otherwise included the same set of potential confounders as the multipollutant models (i.e., temporal trends, meteorologic variables, nonasthma pediatric ED visits for acute upper respiratory infections and interactions between temporal trends and meteorological variables).

Winquist et al.²² observed that joint effect estimates from the multipollutant models were typically smaller than the sum of the estimates from the single-pollutant models that comprised the respective multipollutant groups. As the authors note, this is likely due to unmeasured copollutant confounding in the single-pollutant models. There were not generally notable differences between the joint effect estimates from multipollutant models with and without interaction. The natural logarithm of the RR estimates are used as the $\mathrm{\beta }\mathrm{s}$ in the HIFs (see Excel Table S1 for $\mathrm{\beta }\mathrm{s}$ from all single- and multipollutant models). For each multipollutant model with interaction terms, we extracted the $\mathrm{\beta }\mathrm{s}$ for all interaction terms regardless of statistical significance, because the interaction terms included in the models impact the $\mathrm{\beta }\mathrm{s}$ for the noninteraction terms. We account for the uncertainty in each coefficient by using the variance-covariance matrices in the estimation of CIs.

In addition to the beta coefficient estimates, we extracted the variance-covariance matrices reported in Winquist et al.²² (Excel Table S2) and calculated SEs for the joint effect coefficients for each multipollutant group—i.e., ${\mathrm{\Sigma }}_i({\mathrm{\beta }}_i\times \mathrm{\Delta }A{Q}_i)$. SEs for joint effects were calculated as a function of the variance-covariance matrices of the $\mathrm{\beta }$ estimates and a vector of $\mathrm{\Delta }AQ$ (Equation 4):

$$S{E}_{JE}=\sqrt{L\prime \widehat{\mathrm{\Sigma }}L},$$ (4)

where $S{E}_{JE}$ is the SE of the joint effects for the pollutant group of interest, $\widehat{\mathrm{\Sigma }}$ is the estimated covariance matrix for the pollutant group (see Excel Table S2), and L is the vector of seasonal deltas for each pollutant in the pollutant group, ordered to match $\widehat{\mathrm{\Sigma }}$. The multipollutant version of BenMAP-CE was modified to import variance-covariance matrices and calculate SEs for joint effects using Equation 4. BenMAP-CE uses Monte Carlo simulation to construct CIs around the mean health impact estimates by randomly sampling an uncertainty distribution around the effect coefficients using the SE of the joint effects.

Air Quality Data

Various approaches can be used to develop an air quality delta for use in a HIF. For example, incremental or proportional rollbacks—approaches that are built into BenMAP-CE—are often applied to existing air quality data to create changes in pollutant concentrations. In this analysis, we instead used simulated 2025 air quality projections from 2011 baseline data that were previously developed by the US EPA.²⁵ The included air quality scenarios were chosen here as an illustrative example of a complex change in emissions across a multitude of scenarios. This selection is intended to reflect a more realistic change in air quality rather than simply making incremental or proportional changes to pollutant concentrations.

Air quality data for the years 2011 and 2025 were simulated for the individual air pollutants that comprise each of the air pollutant groups (see Table 1) using the US EPA CMAQ (version 5.1; CMAQv5.1) model.²⁴ Inputs to the model included annual 2011 gridded meteorological fields, emissions data, and boundary conditions. Gridded meteorological fields were provided with the Weather Research and Forecasting (WRF) model²⁶ simulation. Models were applied for a domain covering the conterminous United States and parts of southern Canada and northern Mexico with a $12\times 12\mathrm{km}$ horizontal grid with 25 vertical layers of variable thickness extending up to 50 hPa. Hourly predictions in the lowest model layer, which extended to $\sim 20\mathrm{m}$ above ground, were used to calculate the mean concentrations of air pollutants ($\mathrm{\Delta }\text{AQ}$) in the health impact function. Time and space variant lateral boundary inflow was extracted from a hemispheric scale photochemical model simulation for the same time period.²⁷

Anthropogenic emissions from stationary point, nonpoint stationary (area), and mobile sectors were based on the 2011 National Emissions Inventory.²⁵ Biogenic emissions were estimated with the Biogenic Emission Inventory System (version 3.6.1).²⁸ Daily satellite fire detections were used to generate wildland fire emissions.²⁹ Projected future year anthropogenic emissions representing 2025 were developed using sector specific estimates of growth and anticipated controls. Sector specific economic growth factors and reductions due to specific control plans implemented between 2011 and 2025 were numerous and have been described elsewhere.²⁵ Biogenic, wildland fire, and anthropogenic emissions outside the United States were the same for both 2011 and 2025 model simulations. Point source emissions were input to CMAQ based on coordinate location and nonpoint emissions were gridded to match the CMAQ domain. The 2011 baseline and projected 2025 emissions were developed to support policy assessments and represent a complex representation of growth and control across pollutants, sectors, and regions in the United States to challenge this approach with nonuniform emission reductions.

Hourly measurements made at routine surface monitors (epa.gov/aqs) during 2011 were paired with model predictions in space and time. Normalized mean bias was calculated for all hourly model-observation pairs. Model predictions for ${\mathrm{NO}}_2$ (15% bias), ${\mathrm{O}}_3$ (6% bias), and ${\mathrm{PM}}_{2.5}$ ($-3\%$ bias) compared best with measurements, whereas CO ($-29\%$ bias) and ${\mathrm{SO}}_2$ ($-22\%$ bias) tended to be underestimated.

Baseline Incidence ($Y_0$) and Population (Pop)

HIFs were calculated using county-level incidence rates of pediatric asthma ED visits that were developed using data from the Healthcare Cost and Utilization Project (HCUP).¹⁸ These data, which are preloaded in BenMAP-CE, include incidence rates of pediatric asthma ED visits among children 2–18 y of age, providing a close match to the epidemiological study population. To match the spatial resolution of the air quality data, incidence rates in children were disaggregated from the county level to the 12-km grid level. BenMAP-CE also has built in crosswalks to generate population estimates for 12-km grids using 2010 US census block population data.¹⁸ Population data were generated for children 5–17 y of age to match the age range of the study population from which the epidemiological data were derived.²²

Analyses

All analyses were conducted using a developmental version of BenMAP-CE that recognizes multipollutant groups, accepts air quality surfaces for multiple air pollutants in a single analysis, and processes HIFs of the forms described in Equations 1, 2, and 3. We additionally modified the tool to facilitate the processing of air quality data for multiple air pollutants in a single analysis, updated the calculation of seasonal point estimates and the combination of seasonal estimates, and adjusted the algorithms for estimating uncertainty associated with combining multiple beta coefficients from a multipollutant model (Equation 4). The developmental multipollutant version of BenMAP-CE is publicly available on GitHub.³⁰

We quantified air pollutant–attributable pediatric asthma ED visits using 10 multipollutant functions, including functions with and without interaction terms for each of the five air pollutant groups defined in Table 1. For each analysis, baseline (2011) and control (2025) air quality surfaces for each air pollutant within the air pollutant group being analyzed were loaded into BenMAP-CE. The change in air quality between baseline and control for each air pollutant within each air pollutant group (i.e., $\mathrm{\Delta }A{Q}_i$) was calculated as the average of the daily concentration changes from baseline. The average change in air pollutant concentrations was calculated separately for the cold season (1 November–30 April) and warm season (1 May–31 October) to account for the seasonal C-R functions used in the HIFs. The multipollutant HIFs derived from Winquist et al.²² were applied to the city of Atlanta, consistent with the geographic domain of the study, to calculate counts of pediatric asthma ED visits among children 5–17 y of age (y) attributable to the simultaneous seasonal change in concentrations of air pollutants belonging to one of the source-specific pollutant mixtures (i) from 2011 to 2025 in each 12-km grid cell (j, where J is the total number of grid cells in the city of Atlanta) as (Equation 5):

$$y_i={\sum }_j^J\mathrm{\Delta }{y}_{ij}.$$ (5)

$\mathrm{\Delta }{y}_{ij}$ was estimated using Equations 2 (no interaction) and 3 (first-order interactions), as described above. In addition, $y_i$ was estimated separately for the warm and cold season, accounting for the seasonal air pollution deltas and beta coefficients, and an annual estimate was calculated by summing the seasonal results. The delta method was used to derive the variance of the summed estimates and construct 95% CIs.

To compare these results to analyses that more commonly focus on a single pollutant, we additionally analyzed single-pollutant HIFs for the individual pollutants (k; where K is the total number of pollutants in a given source-specific pollutant mixture) that constitute each of the pollutant groups and summed them to approximate the counts of pediatric asthma ED visits attributable to changes in different source-specific pollutant mixtures (Equation 6):

$$y_i={\sum }_j^J{\sum }_k^K\mathrm{\Delta }{y}_{kj}.$$ (6)

As a sensitivity analysis to examine the application of these approaches to a larger spatial domain, the same multipollutant and single-pollutant analyses were conducted for the entire state of Georgia.

Results

Predicted Changes in Air Quality

Based on the estimated growth and anticipated controls employed on air pollution sources, CMAQ predicted decreases in most of the modeled pollutants in 2025 in comparison with 2011 [Table 3 (Atlanta) and Table S1 (Georgia)]. The largest relative decrements in annual average pollutant concentrations in Atlanta were predicted for EC ($-55.1\%$; $-0.9{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$), ${\mathrm{SO}}_2$ ($-52.5\%$; $-1.4\text{\hspace{0.17em}ppb}$), ${\mathrm{NO}}_2$ ($-45.2\%$; $-17.0\text{\hspace{0.17em}ppb}$), and secondary ${\mathrm{PM}}_{2.5}$ ($-40.0\%$; $-1.7{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$). Minimal changes in concentrations were predicted for ${\mathrm{O}}_3$ ($-1.4\%$; $-0.6\text{\hspace{0.17em}ppb}$). Notably, relative predicted changes in seasonal averages were comparable for all pollutants except ${\mathrm{O}}_3$, for which CMAQ predicted an increase in cold season average (10.7%; $3.6\text{\hspace{0.17em}ppb}$) and a decrease in warm season average ($-8.7\%$; $-4.5\text{\hspace{0.17em}ppb}$) concentrations.

Table 3.

Predicted changes in seasonal and annual pollutant concentrations in Atlanta, Georgia, from 2011 to 2025, using US EPA Community Multiscale Air Quality (version 5.1; CMAQv5.1) model.

Pollutant	Annual			Warm season			Cold season
	2011	2025	$\mathrm{\Delta }$ (%)	2011	2025	$\mathrm{\Delta }$ (%)	2011	2025	$\mathrm{\Delta }$ (%)
CO (ppm)	0.6	0.5	$-0.2$ ($-25.9$)	0.7	0.5	$-0.2$ ($-23.7$)	0.6	0.5	$-0.2$ ($-28.1$)
EC (${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$)	1.6	0.7	$-0.9$ ($-55.1$)	1.5	0.6	$-0.9$ ($-58.1$)	1.6	0.8	$-0.8$ ($-52.1$)
${\mathrm{NO}}_2$ (ppb)	37.6	20.6	$-17.0$ ($-45.2$)	39.6	19.4	$-20.2$ ($-50.9$)	35.5	21.8	$-13.7$ ($-38.6$)
${\mathrm{O}}_3$ (ppb)	43.7	43.1	$-0.6$ ($-1.4$)	54.0	49.3	$-4.7$ ($-8.7$)	33.3	36.9	3.6 ($-10.7$)
${\mathrm{PM}}_{2.5}$ (${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$)	12.8	10.0	$-2.9$ ($-22.4$)	12.5	9.3	$-3.2$ ($-25.8$)	13.2	10.6	$-2.5$ ($-19.1$)
Secondary ${\mathrm{PM}}_{2.5}$ (${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$)	4.2	2.5	$-1.7$ ($-40.0$)	3.8	1.9	$-1.9$ ($-50.8$)	4.6	3.2	$-1.4$ ($-30.9$)
${\mathrm{SO}}_2$ (ppb)	2.7	1.3	$-1.4$ ($-52.5$)	2.9	1.2	$-1.7$ ($-57.7$)	2.5	1.4	$-1.2$ ($-46.2$)
${\mathrm{SO}}_4$ (${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$)	2.2	1.4	$-0.8$ ($-37.5$)	2.6	1.4	$-1.2$ ($-46.6$)	1.8	1.4	$-0.4$ ($-24.1$)

Note: CO, carbon monoxide; Cold season, November–April; EC, elemental carbon; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{O}}_3$, ozone; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; ppb, parts per billion; ppm, parts per million; ${\mathrm{SO}}_2$, sulfur dioxide; ${\mathrm{SO}}_4$, sulfate; Warm season, May–March.

Single-Pollutant Results

Results from single-pollutant models varied considerably by pollutant and season (Figure 1; Table 4). For the following results, avoided ED visits in the warm and cold seasons represent an estimated decrease in ED visits attributable to a change in seasonal air pollution concentrations in 2025 in comparison with 2011. The estimated number of avoided pediatric asthma ED visits in Atlanta attributable to the modeled changes in single pollutants were consistently higher in the warm season in comparison with the cold season. Individual estimates ranged from 2.4 (95% CI: 1.6, 3.2; ${\mathrm{SO}}_2$) to 42.7 (95% CI: 36.3, 49.1; ${\mathrm{NO}}_2$) visits in the warm season and from $-2.8$ (95% CI: $-6.3$, 0.6; ${\mathrm{O}}_3$) to 7.0 (95% CI: 1.6, 12.5; ${\mathrm{NO}}_2$) in the cold season, where positive numbers represent avoided ED visits and negative numbers represent an estimated increase in ED visits. We estimated that predicted changes in annual ${\mathrm{NO}}_2$ concentrations were responsible for the most avoided annual pediatric asthma ED visits of any pollutant [49.7 (95% CI: 41.3, 58.2)], accounting for more than twice as many as the pollutant responsible for the second most avoided ED visits [EC: 18.3 (95% CI: 10.9, 25.8)]. The estimated numbers of avoided pediatric asthma ED visits across the entire state of Georgia were notably higher than those in Atlanta due to the increased population total, but the seasonal and between-pollutant patterns are comparable (Figure S1 and Table S2).

Figure 1.

Estimated number of annual and seasonal avoided pediatric asthma emergency department visits (5–17 y of age) in 2025 due to predicted changes in air quality in Atlanta, Georgia, from 2011 to 2025, quantified using single-pollutant and multipollutant health impact functions in a developmental version of BenMAP-CE. Note: Negative numbers represent a predicted increase in asthma ED visits. In addition to being identified via color, health impact function model types are listed in the order that they appear in each pollutant group/season grid box. Quantitative summary data presented in Tables 4–7. Dots on figure represent point estimates for avoided asthma ED visits and lines represent 95% confidence intervals. BenMAP-CE, Environmental Benefits Mapping and Analysis Program–Community Edition; CO, carbon monoxide, Cold season, November–April; Criteria pollutants, CO, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, ${\mathrm{PM}}_{2.5}$, and ${\mathrm{SO}}_2$; EC, elemental carbon; ED, emergency department; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{O}}_3$, ozone; Oxidants, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, and ${\mathrm{SO}}_2$; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; Power plant, ${\mathrm{SO}}_2$ and sulfate; Secondary pollutants, ${\mathrm{O}}_3$ and secondary ${\mathrm{PM}}_{2.5}$; ${\mathrm{SO}}_2$, sulfur dioxide; ${\mathrm{SO}}_4$, sulfate; Traffic pollutants, CO, EC, and ${\mathrm{NO}}_2$; Warm season, May–March.

Table 4.

Estimated number of annual and seasonal avoided pediatric asthma emergency department visits (5–17 y old) in 2025 due to predicted changes in air quality in Atlanta, Georgia, from 2011 to 2025, quantified using single-pollutant health impact functions in BenMAP-CE.

Pollutant	Annual	Cold season	Warm season
	PE (95% CI)	PE (95% CI)	PE (95% CI)
CO	9.4 (7.1, 11.7)	1.4 ($-0.6$, 3.4)	8.0 (6.9, 9.2)
EC	18.3 (10.9, 25.8)	1.3 ($-4.6$, 7.2)	17.0 (12.5, 21.6)
${\mathrm{NO}}_2$	49.7 (41.3, 58.2)	7.0 (1.6, 12.5)	42.7 (36.3, 49.1)
${\mathrm{O}}_3$	1.4 ($-2.4$, 5.2)	$-2.8$ ($-6.3$, 0.6)	4.2 (2.6, 5.9)
${\mathrm{PM}}_{2.5}$	3.1 (1.7, 4.6)	0.6 ($-0.6$, 1.8)	2.5 (1.8, 3.3)
Secondary ${\mathrm{PM}}_{2.5}$	2.1 ($-0.2$, 4.4)	$-1.8$ ($-3.5$, $-0.1$)	3.8 (2.3, 5.4)
${\mathrm{SO}}_2$	1.8 (0.8, 2.7)	$-0.6$ ($-1.2$, 0.0)	2.4 (1.6, 3.2)
${\mathrm{SO}}_4$	2.8 (0.4, 5.1)	$-0.4$ ($-0.9$, 0.1)	3.2 (0.9, 5.4)

Note: Negative numbers represent a predicted increase in asthma emergency department visits. BenMAP-CE, Environmental Benefits and Mapping Program–Community Edition; CI, confidence interval; CO, carbon monoxide; Cold season, November–April; EC, elemental carbon; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{O}}_3$, ozone; PE, point estimate; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; ${\mathrm{SO}}_2$, sulfur dioxide; ${\mathrm{SO}}_4$, sulfate; Warm season, May–March.

Multipollutant and Sum of Single-Pollutant Results

Multipollutant results for Atlanta exhibit some of the same trends displayed in the single-pollutant results (Figure 1; Tables 5–7). Specifically, estimated changes in avoided pediatric asthma ED visits attributable to simultaneous changes in multiple pollutants, as defined by the specified pollutant groups, were driven mostly by avoided pediatric asthma ED visits in the warm season. In addition, we estimated the greatest number of avoided pediatric asthma ED visits for pollutant groups that include ${\mathrm{NO}}_2$ (i.e., criteria pollutants, oxidants, and traffic pollutants). In models that did not account for potential interaction, year-round estimates for the ${\mathrm{NO}}_2\text{-inclusive}$ pollutant groups ranged from 32.4 (95% CI: 19.3, 45.5; traffic pollutants) to 42.6 (95% CI: 31.9, 53.3; oxidants) avoided pediatric asthma ED visits, in comparison with 0.1 (95% CI: $-2.6$, 2.8; secondary pollutants) and 3.7 (95% CI: 2.0, 5.4; power plant pollutants) for the groups that did not include ${\mathrm{NO}}_2$. In general, the estimated annual impacts on ED visits attributable to any given pollutant group were greater (i.e., farther from the null) when estimated using C-R parameters that included interaction terms. The one exception to this is the traffic pollutants group, which had interaction-based estimates that were comparable to, but less precise than, the estimates without interaction terms. Notably, this exception did not hold for the results including the entire state of Georgia, which otherwise demonstrated patterns similar to those of the Atlanta results (Figure S1 and Table S3–S5).

Table 5.

Estimated number of annual and seasonal avoided pediatric asthma emergency department visits (5–17 y old) in 2025 due to predicted changes in air quality in Atlanta, Georgia, from 2011 to 2025, quantified using multipollutant health impact functions without interaction terms in a developmental version of BenMAP-CE.

	Joint effects without interaction terms
	Annual	Cold season	Warm season
Pollutant group	PE (95% CI)	PE (95% CI)	PE (95% CI)
Criteria	32.8 (18.9, 46.6)	5.9 ($-2.1$, 13.8)	26.9 (15.5, 38.2)
Oxidants	42.6 (31.9, 53.3)	2.7 ($-3.2$, 8.5)	40.0 (31.0, 48.9)
Power plant	3.7 (2.0, 5.4)	$-0.8$ ($-1.6$, 0.1)	4.5 (3.0, 5.9)
Secondary	0.1 ($-2.6$, 2.8)	$-5.1$ ($-7.2$, $-2.9$)	5.2 (3.5, 6.8)
Traffic	32.4 (19.3, 45.5)	7.7 (0.4, 14.9)	24.7 (13.8, 35.6)

Note: Negative numbers represent a predicted increase in asthma emergency department visits. BenMAP-CE, Environmental Benefits and Mapping Program–Community Edition; CI, confidence interval; CO, carbon monoxide; Cold season, November–April; Criteria pollutants, CO, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, ${\mathrm{PM}}_{2.5}$, and ${\mathrm{SO}}_2$; EC, elemental carbon; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{O}}_3$, ozone; Oxidants, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, and ${\mathrm{SO}}_2$; PE, point estimate; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; Power plant, ${\mathrm{SO}}_2$ and sulfate; Secondary pollutants, ${\mathrm{O}}_3$ and secondary ${\mathrm{PM}}_{2.5}$; ${\mathrm{SO}}_2$, sulfur dioxide; ${\mathrm{SO}}_4$, sulfate; Traffic pollutants, CO, EC, and ${\mathrm{NO}}_2$; Warm season, May–March.

Table 7.

Estimated number of annual and seasonal avoided pediatric asthma emergency department visits (5–17 y old) in 2025 due to predicted changes in air quality in Atlanta, Georgia, from 2011 to 2025, quantified using summed results from single-pollutant health impact functions in BenMAP-CE.

	Sum of single pollutants
	Annual	Cold season	Warm season
Pollutant group	PE (95% CI)	PE (95% CI)	PE (95% CI)
Criteria	65.5 (55.8, 75.2)	5.6 ($-1.3$, 12.5)	59.9 (53.1, 66.7)
Oxidants	52.9 (43.6, 62.2)	3.6 ($-2.9$, 10.1)	49.3 (42.6, 56.0)
Power plant	4.5 (2.0, 7.0)	$-1.0$ ($-1.8$, $-0.3$)	5.5 (3.2, 7.9)
Secondary	3.5 ($-1.0$, 7.9)	$-4.6$ ($-8.4$, $-0.8$)	8.1 (5.8, 10.3)
Traffic	77.5 (66.0, 89.0)	9.7 (1.4, 18.0)	67.8 (59.8, 75.8)

Note: Negative numbers represent a predicted increase in asthma emergency department visits. BenMAP-CE, Environmental Benefits and Mapping Program–Community Edition; CI, confidence interval; CO, carbon monoxide; Cold season, November–April; Criteria pollutants, CO, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, ${\mathrm{PM}}_{2.5}$, and ${\mathrm{SO}}_2$; EC, elemental carbon; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{O}}_3$ ozone; Oxidants, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, and ${\mathrm{SO}}_2$; PE, point estimate; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; Power plant, ${\mathrm{SO}}_2$ and sulfate; Secondary pollutants, ${\mathrm{O}}_3$ and secondary ${\mathrm{PM}}_{2.5}$; ${\mathrm{SO}}_2$, sulfur dioxide; ${\mathrm{SO}}_4$, sulfate; Traffic pollutants, CO, EC, and ${\mathrm{NO}}_2$; Warm season, May–March.

Table 6.

Estimated number of annual and seasonal avoided pediatric asthma emergency department visits (5–17 y old) in 2025 due to predicted changes in air quality in Atlanta, Georgia, from 2011 to 2025; quantified using multipollutant health impact functions with interaction terms in a developmental version of BenMAP-CE.

	Joint effects with interaction terms
	Annual	Cold season	Warm season
Pollutant group	PE (95% CI)	PE (95% CI)	PE (95% CI)
Criteria	45.8 (24.8, 66.7)	4.6 ($-7.2$, 16.5)	41.1 (23.9, 58.4)
Oxidants	55.4 (41.8, 69.0)	12.4 (3.3, 21.5)	43.0 (32.9, 53.1)
Power plant	7.3 (2.9, 11.6)	$-2.6$ ($-4.0$, $-1.1$)	9.8 (5.8, 13.9)
Secondary	$-1.1$ ($-4.0$, 1.8)	$-6.3$ ($-8.5$, $-4.1$)	5.2 (3.3, 7.2)
Traffic	27.1 (1.6, 52.7)	9.4 ($-1.3$, 20.1)	17.7 ($-5.4$, 40.9)

Note: Negative numbers represent a predicted increase in asthma emergency department visits. BenMAP-CE, Environmental Benefits and Mapping Program–Community Edition; CI, confidence interval; CO, carbon monoxide; Cold season, November–April; Criteria pollutants, CO, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, ${\mathrm{PM}}_{2.5}$, and ${\mathrm{SO}}_2$; EC, elemental carbon; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{O}}_3$, ozone; Oxidants, ${\mathrm{NO}}_2$, ${\mathrm{O}}_3$, and ${\mathrm{SO}}_2$; PE, point estimate; ${\mathrm{PM}}_{2.5}$, particulate matter with a nominal mean aerodynamic diameter less than or equal to $2.5\mathrm{\mu }\mathrm{m}$; Power plant, ${\mathrm{SO}}_2$ and sulfate; Secondary pollutants, ${\mathrm{O}}_3$ and secondary ${\mathrm{PM}}_{2.5}$; ${\mathrm{SO}}_2$, sulfur dioxide; ${\mathrm{SO}}_4$, sulfate; Traffic pollutants, CO, EC, and ${\mathrm{NO}}_2$; Warm season, May–March.

The results from the multipollutant analyses were compared with analyses that summed the single-pollutant results for pollutants that comprised each pollutant group. A comparison of these results is presented in Figure 1 and Tables 5–7. Although there was some indication of variability in results between the sum of single-pollutant and multipollutant (both with and without interaction) analyses, across each category the results were relatively similar (i.e., overlapping 95% CIs) for all categories except traffic and criteria pollutants. On average, the changes in pediatric asthma ED visits in the cold season estimated from multipollutant HIFs with and without interaction terms were within one estimated ED visit of the corresponding sums of the single-pollutant results for each of the pollutant groups. For the traffic and criteria pollutant categories in the warm season, the sums of single-pollutant estimates were substantially larger than both multipollutant analyses (46%–383%).

Discussion

In this case study, we assessed the utility and feasibility of implementing a multipollutant approach to estimating the health impacts associated with changes in air quality. In the multipollutant analyses, we found that the estimated number of avoided pediatric asthma ED visits for the source categories examined were relatively similar, though the interaction model often estimated larger impacts among the source categories. The estimated multipollutant results were generally of the same magnitude as the summed single-pollutant results; the exceptions being the results for the traffic and criteria pollutants in the warm season. The relative consistency of results in the computationally intensive multipollutant approach in comparison with the simpler sum of the single-pollutant results indicates that it is worthwhile to consider the complexity of the statistical approach used to estimate multipollutant impacts along with the time and resources available to institute a multipollutant approach in the process of planning future quantitative assessments. Ultimately, the value added by the multipollutant approach in comparison with the single-pollutant approach depends on both on the shape of C-R relationships and nature of the air quality data used in the analysis. A strength of this work is that we demonstrated the capabilities of a publicly available developmental version of BenMAP-CE with multipollutant functionality. This developmental version of BenMAP-CE reduces the initial investment necessary to other researchers for further exploratory health impact analyses using multipollutant effect estimates from additive main effects models.

Although we successfully implemented a multipollutant approach to estimating health impacts within BenMAP-CE, we also identified several challenges inherent in estimating the health impacts associated with changes in air quality within a multipollutant context. Most important, there is a fundamental question as to whether there is scientific evidence to support the use of results from multipollutant models in the quantification of the health impacts associated with changes in air quality. In estimating health impacts due to changes in air quality, ideally there would be some degree of confidence in the air pollutant–health effects relationship prior to conducting such an assessment. Confidence in the air pollutant–health effects relationship directly translates to confidence in the estimated number of health impacts due to changes in air quality. Without scientific evidence supporting a relationship between an air pollutant exposure and a health effect, there is significant uncertainty in the estimated number of health impacts caused by changes in air quality. As a result, standard US EPA practice is to only quantify impacts for pollutants where the scientific evidence supports some degree of certainty regarding the causal nature of the relationship between the air pollutant exposure and health outcome of interest¹⁸ (i.e., likely to be causal relationship or causal relationship).

Although there has been limited research to date examining relationships between short-term multipollutant exposures and health effects, for the purposes of this study we applied multipollutant effect estimates from a study that focused on the outcome of asthma ED visits because there is extensive scientific evidence that supports a “likely to be causal” or “causal” relationship between short-term exposures to some of the individual criteria pollutants examined (i.e., ${\mathrm{SO}}_2$, ${\mathrm{NO}}_2$, ozone, and ${\mathrm{PM}}_{2.5}$) and respiratory effects, including asthma. Specifically, the most recent US EPA Integrated Science Assessments (ISAs) for the criteria pollutants concluded that the evidence for short-term exposures and respiratory effects supports a “causal relationship” for ozone,³¹ ${\mathrm{NO}}_2$,³² and ${\mathrm{SO}}_2$³³; “likely to be causal relationship” for ${\mathrm{PM}}_{2.5}$³⁴; and “suggestive of, but not sufficient to infer, causal relationship” for CO.³⁵ We highlight the issue of causality, because quantifying multipollutant impacts by estimating joint effects for a group of air pollutants with strong individual evidence of causality is defensible, given that the beta coefficients used in the health impact functions represent the estimated independent effect of each of the air pollutants while controlling for the potential confounding effects of the other air pollutants in the model. However, there is more uncertainty in the interpretation of health impacts quantified using joint effects from models with interaction terms, given that systematic evaluations that are on par with ISAs have not been conducted to assess causality between two or more air pollutant exposures that interact to affect health outcomes on a nonadditive scale.

In addition to concerns about the appropriateness of quantifying the health impacts of multipollutant exposures because of uncertainties regarding the strength of the relationship between specific combinations of pollutants and health, the ability to estimate the potential health impacts of multipollutant exposures is limited by the statistical approach used. To date, although numerous epidemiological studies have employed various statistical approaches to examine associations between short-term air pollution exposures and health,¹⁷ the complexity and differences in statistical approaches employed complicates the interpretation of results across studies.¹⁶ As a result, in this case study, we chose to apply effect estimates from a study that used additive main effects models because they provide a less computationally intensive method to assess the health effects of multipollutant exposures in comparison with other methods, such as dimension reduction, nonparametric models, and novel multipollutant models.¹⁶ As we experienced developing this multipollutant version of BenMAP-CE, even a less complex method required a significant level of effort to implement in a health impact assessment. Further, some more complex models may not be directly translatable to a parameterized health impact function. Despite their relative ease of computation and interpretability, additive main effects models may produce unstable effect estimates due to multicollinearity. In addition, individual pollutant effect estimates within an additive main effects model may vary due to differential exposure measurement error. The variances of the beta coefficients used in this analysis were not severely inflated,²² suggesting that the impact of collinearity was not excessive.³⁶ The developmental multipollutant version of BenMAP-CE could also accommodate joint effect estimates from other multipollutant modeling approaches that may more adequately estimate joint effects from highly correlated exposures, such as parametric g-computation.³⁷ However, we recognize that we introduced uncertainty in this analysis by applying potentially unstable individual pollutant effect estimates that comprised the joint effects (see Equations 2 and 3).

Health impact analyses often use effect estimates from meta-analyses or several individual studies that represent a range of estimates from the larger literature base.³⁸ Although literature on the relationship between multipollutant exposures and health effects is expanding, there is limited evidence for additive main effects models that evaluate the same pollutant groups over the same geographic domain. Thus, this analysis relies on multipollutant effect estimates from a single study, which raises uncertainty regarding the robustness of the selected effect estimates. Many of the effect estimates used in this analysis are imprecise, and some are indicative of protective effects (see Excel Table S1), which further highlights this uncertainty. We selected effect estimates from this particular study because it was a large time-series study conducted in a single city. The length of the time series was likely to produce more stable effect estimates,³⁹ and the limited geographic scale reduced the processing power required by BenMAP-CE to load and manipulate air quality surfaces for multiple pollutants. To reduce uncertainties related to generalizability, the geographic domain of this case study was chosen to match the location of the epidemiological study from which we extracted effect estimates. Given the spatiotemporal variability of pollutants and their relationships to each other,^40,41 as well as the population attributes specific to the epidemiological study location, the results are not likely generalizable to other locations. The analysis described in this paper demonstrates the feasibility of estimating multipollutant health impacts in BenMAP-CE and provides a novel comparison between single- and multipollutant air pollutant estimates, but the absolute health impacts estimated should be interpreted with caution due to the noted uncertainties and lack of generalizability to other geographic locations.

Conclusion

In this case study, we demonstrated that a multipollutant approach to estimating the health impacts of changes in air pollution is feasible but incurs a number of challenges unique to multipollutant assessments. Despite incorporating a less complex statistical approach to estimating the health impacts of multipollutant exposures, there was a substantial investment of time and resources to produce a functional multipollutant version of BenMAP-CE. In this analysis, the sums of the individual results were generally consistent with the multipollutant results, which raises questions about the value added relative to the time and resource investment. Notably this analysis generated results from one multipollutant model type applied in one city. Thus, further research is warranted to reach more definitive conclusions about the potential added value for any particular multipollutant methodology. However, although the state of multipollutant research has advanced substantially over the past decade, the lack of consistent and comparable methods across studies makes it difficult to confidently conduct risk assessments for multipollutant exposures. In summary, this work shows that although it is possible to use a multipollutant approach to estimate health impacts associated with changes in multiple air pollutants on a large scale (e.g., at the city and state level), the development of a multipollutant version of BenMAP-CE was computationally complex and incurred large time and resource costs. The comparability of results generated from single-pollutant and multipollutant effect estimates indicates that the trade-offs between single-pollutant and multipollutant approaches should be considered in the planning of future quantitative assessments.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.